Here's a test run after I painted the hole (the side walls of the hole). I upped the filtering (Savitzky-Golay) to 10 seconds from five. In hind sight, I think that was a mistake, because when you're trying to figure out what the heck is going on, you probably don't want something hiding any data from you.



For the next test I made spray painted (hi temp flack black) a little strip of copper foil and rolled into a tube, which I pressed into and through the hole. It was probably 3/8 long, so should have blocked the sensor from seeing anything but the drum and itself. During this test, I pulled the drum away from the burner and shot it with an handeld IR gun to get an actual temp. At the end when I turned the gas off, I tipped the drum up so I could take a couple of actual readings as things cooled down. I also added a sensors - a probe insert an inch or two above the IR sensor, through the wall and less than an inch from the drum itself.



There still seems to be quite a correlation between "flame" and the drum IR reading. When the flame is turned on, it shoots up, and when the flame is turned off, it shoots down. With the flame off, the analog probe near the drum seems to track pretty well with actual drum temp, but with flame on, it also runs much hotter than the drum.

It was interesting that this time when I pushed the sensor in, it registered hotter. When I did this same test without the tube, it brought the reading cooler.

With the painted drum, I notice my manual IR gun readings are much more consistent and repeatable. I think they were suffering from reflection issues previously.

Bottom line, adding the view tube doesn't seem like it really gained me anything. I kinda want to go back and redo the pained wall test again without the data filtering, and with the occasional manual temp checks to see if I can pick up any correlations/observations.

I still think I'm relying on the "in the mail" narrow field of view sensor to make this work.

Doh! I took a close look at the sensor and found that I somehow had completely painted it black. I think it was when I painted the drum, though I swear I had everything taped off. Oh well... so I think the tests I've done since painting the drum are likely all been with an painted sensor, so it was likely just seeing the temperature on it's opaque surface, not the drum at all. It cleaned up ok, and it really felt like the hi temp paint and not soot or smoke accumulation.

Here's a rerun of the sensor peering through a hole in the roaster wall. Periodically, I pulled the drum away from the heat so I could shoot a handheld IR temp - you can see the dips in the drum temp line when I did that. I did that a handful of times, and then just plotted a straight line between the points to approximate an "actual drum temp". Clearly, there's sill some flame related inflation happening that's kicking up the IR sensor drum temp. When I take the flame away, it's starts to get close to accurate Either somehow there's reflection being picked up or the wave of heat related to the burner that hits the sensor is mucking up it's difference calculation. I think I'll hang it up with this sensor for now and wait until the other arrives.

I think Version 2 is going to be a winner

The new 5 degree field of view sensor can be further away, so I did a new larger bracket for it, and also incorporated a larger 40mm fan. Being further away, I was a little worried about aiming it, so I also put in an aiming capability (the 4 screws that hold the sensor can be independently lengthened or shortened). Here's what it looks like:



With the 90 degree FOV sensor, I would see 25 to 50 degree jumps whenever I turned the flame up, so somehow it was reacting to seeing the actual flame or the wave of heat. With the new sensor, there is very little jump when changing gas. Similar to the very first test I ran at the beginning of this thread, here's an empty drum turning with varying levels of gas:



The sensor stays way cooler (the IRT temp I recorded) with the new fan and being further away, so that's going to help a lot as well. Don't see any issues keeping it under it's 125C limit.

Another observation is that the raw readings from the sensor look "jumpy" - the kind of oscillate up and down over a few degree range. I notice when I turn the drum off the line smooths out - so the oscillations correlate to some kind of variation in the drum surface, or maybe even actual heat variations due to where the vanes are welded, etc.). It's not a big deal.

Now to do a real roast and see how useful this is going to be....

Nice.
You may also want to consider a cap when not in use. I have a plastic cap that I install on the sensor when not in use. All you need is a tiny dust bunny to cause very strange results. This has never happened to me, but I am on a lookout for an occurrence.

Ugly roast aside... ran into an issue where the DT seems to max out at 292C (557F), when it's clearly getting hotter. It's not a hard limit, as it will vary a few tenths from second to second. It's not Artisan, as I can see the temp when it leaves the Arduino (before it gets to Artisan) and it's also limited there. So it's either in the Arduino or in the device itself - I know you can configure the range of these devices, and I haven't looked yet to see the range it was set up with when I received it. Right now I have it transmitting as a short integer (2 byte), with 1/100 multiplier, so 250 degrees would be 25000. Given the drum is getting up into at least the mid 300C range, I think I have to knock that back to a 1/10 multiplier (250 degrees being 2500), so that I stay under the magic 32768 limit for a short integer. I can't imagine needing the DT in 100's of a degree vs tenths.

I've very happy with the accuracy up to that limit. The sensor itself (IRT) stayed nice and cool with the new "further away" bracket and 40mm fan. The idea of a cap is a good idea to guard for dust bunnies is a good idea.

BTW, I forgot to turn on the gas at the cylinder, so was only running on gas that was left in the line initially. When I noticed the flame dipping, I turned it on - that's why the drum temp started petering out at 3 minutes.

I noticed when looking at the fan mounting that it appears to me like you're covered up much of the inlet to the fan with the mounting brackets. You might find dramatic increases in airflow if you open up the inlet side.

Ira

After taking a deep dive into what's happening with the sensor as I hit 290 degrees... I see the internal raw IR reading (pre-scaling) peg at 0x'7FFF', so I don't think it's an internal scaling issue.

What I have in place at the moment is a MLX90614ESF-DCI sensor. That's a 5 degree field of view, with extra accuracy for medical purposes. Sounds like a specialty item, but maybe with COVID they've become surplus and easy to get your hands on. Unfortunately, I think the pegging at 290 degrees is due to the "extra accuracy for medical purposes". I've not been able to find good documentation on exactly what this limit is, but I did find a blurb that said the special amplifier used in the medical grade sensor will saturate around 200 degrees. And I find a another cryptic note that the max temp on the medical grade sensor is reduced. "Saturation" is a good description for the pegging I'm seeing. I guess if you've got a fever over 200 degrees C you're a goner, anyway.

What I want, I think, is the MLX90614ESF-BCI, which is the same 3V, 5 degree field of view sensor, but without the medical accuracy/reduced range. That hopefully, will go up to 380 degrees, as advertised. I should be able to swap that in without any changes to the bracket and mounting. Yeah, a little unsoldering, re-soldering, but not too bad. ​

Regarding the fan, I think the comment about more airflow with a "more clearance" mounting is correct. As long as the temp stays down, I'll probably leave it as is. I stuffed the 40mm fan in there partly so I'd have air to spare.

Fyi - here's a closeup of the intake side of the fan:

Confirmed... the MLX90614ESF-BCI sensor is the one to get. I swapped one of those into my setup and tested it all the way up to 380. When it gets to 380 it just doesn't read any higher. But I think 380 if plenty. Here's my first roast with this sensor:



The gas is automated and I'm working on my algorithm preceding and during first crack... more work to be done there